TL;DR
Australia's autonomous Ghost Bat jet fired and destroyed an aerial target with a missile in a desert test late last year, marking a concrete milestone in AI-assisted lethal autonomy and accelerating a global race to field combat AI.
What happened
- Ghost Bat, an Australian-made unmanned jet roughly two-thirds the size of a conventional fighter, shot down an aerial target with an air-to-air missile over South Australia's remote desert before Christmas 2025.
- The aircraft is not AI-piloted in the generative sense: Air Marshal Stephen Chappell confirmed it runs on deterministic programming, choosing from thousands of pre-designed options validated through digital-twin and AI-assisted simulation.
- An operator aboard a E7 Wedgetail aircraft ordered the Ghost Bat to engage, the system selected its own attack plan, and the operator gave final launch authorisation before the missile fired.
- Defence Minister Richard Marles separately traveled to the United States to meet OpenAI, Anthropic, Microsoft, and Google on where their models are heading for defence applications.
- The ADF has also adopted Palantir's Maven Smart system for intelligence analysis, drawing controversy over founder political ties and data sovereignty concerns.
Why it matters
- The test proves human-machine teaming for lethal strikes is operational, not theoretical: a sovereign Australian platform can autonomously plan and execute an engagement with a human authorising only the final step.
- Deterministic architecture is a deliberate legal and ethical choice: by constraining the Ghost Bat to pre-validated options, the ADF preserves predictability and legal accountability, a model other militaries will study.
- The milestone lands as AI-guided armed drones are already active in Ukraine, with Ukrainian forces identifying Russian drones carrying minicomputers and no antennas as almost certainly AI-piloted, and Ukraine running its own AI drones against Russian supply routes.
- Israel's use of AI targeting tools, including the reported Lavender system that generated a list of 37,000 suspected operatives as potential strike targets, has drawn sharp scrutiny from the IISS, which warns faster targeting increases error risk.
- The UN and ICRC issued a joint call to ban autonomous weapons that are unpredictable or that target humans, citing unacceptable civilian harm in technologically sophisticated conflicts.
What to watch next
- Whether other Five Eyes or NATO partners adopt the deterministic-plus-human-authorisation model as a de facto legal standard, or whether operational pressure pushes toward faster, less supervised autonomy.
- Progress on the Ghost Shark autonomous submarine, the Ghost Bat's underwater counterpart, as the next visible test of ADF autonomous lethality.
- Outcomes of Marles's conversations with OpenAI, Anthropic, Microsoft, and Google: any announced contracts or capability integrations would signal how quickly large-model AI moves from intelligence analysis into operational command support.
Originally published on Present of AI, a daily source-linked AI news timeline. Read the full timeline or browse the open dataset.